Compatibility with LLama

[wj210]

Hi, thanks for this great work.

Like to ask for the counterfactual generation, I understand that the prediction model is first trained on the dataset before using it to generate class labels.

Could I instead use a pretrained model such as LLama?
In this case, what form of output is required from the model? Could I just use model.generate?

Also in the model_specific_predictions_explanations, explanation_model and prediction model are not passed in.